Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 101 === This thesis mainly discusses the design and implementation of object recognition and visual perception for home service robot. In the beginning, the object recognition system built by combining Compute Unified Device Architecture (CUDA) Speeded Up Robust Features (SURF) detector and Binary Robust Invariant Scalable Keypoints (BRISK) descriptor is presented. CUDA SURF improves the speed of original SURF through Graphic Processing Unit (GPU) parallelization. Unlike traditional descriptors, BRISK uses the binary string instead of floating vector to store the description. It not only decreases the consumption on memory but also reduces the computation time for matching. Next, we propose a visual perception system for searching objects, which are not included in database. By calculating the depth difference between two sides of an edge, a probability weight is generated to evaluate whether it is a boundary or not. The contours of objects are searched according to the probability. Then, the Human-Computer Interaction (HCI) is introduced in two ways: for developers, these interfaces should be clear and easy to understand, and for operators, the designs of speech recognition system and gesture recognition system make it convenient to convey instruction or get the attention from robot. In the end, these methods mentioned above are implemented in the laboratory and the competition in robot@home league at RoboCup Japan Open 2013 Tokyo, and the validity and practicability of this real-time object recognition system are demonstrated.
|